What i found is only CanESM2 (for CMIP5) can be used for downscaling using SDSM as SDSM needs predictor variables of the same GCM which is available only for CanESM2.
SDSM is a user-friendly software package designed to implement statistical downscaling methods to produce high-resolution monthly climate information from coarse-resolution climate model (GCM) simulations.
I work with SDSM and Lars-wg5.5. I think in SDSM you can't use it for CMIP5 output and downscaling them. I think you can just use one of Canadian models because your model should has same resolution with NCEP and it is limitation of SDSM. In SDSM you can run it step by step in a clear way and you can't add any others files for downscaling. In a nutshell, you prepare observation files and after that select your aim model and scenario and you can't added any other models of CMIP5.
So, it is better for you to use CORDEX, but in CORDEX you have a problem with spatial resolution. As you told, the best resolution of it, is 0.42*0.42 degree and it is a coarse resolution.
I suggest to you, use of other statistical downscaling methods, such as quantile mapping and other the bias correction methods. There are different packages and cods that can help to you for this way.
What i found is only CanESM2 (for CMIP5) can be used for downscaling using SDSM as SDSM needs predictor variables of the same GCM which is available only for CanESM2.